← All tools

Idea Assessor

An instrument that forces judgment on ideas.

AI has made refining an idea free. Judging one has not become any easier. Every model release makes it cheaper to reword the pitch one more time, and the moment of deciding, or discarding, never arrives. Idea Assessor puts that decision on a clock.

Deferred judgment that looks like work is Cognitive Debt in its most everyday form. The constraint is the product: without a deadline every iteration is free, so none of them count.
Five rounds, then a verdict. The round budget and the kill rule are not UX details, they are the product.

Every idea runs through a fixed YC-style rubric: Problem, Market and Timing, Solution, Moat, Distribution. You get exactly five rounds to sharpen the weakest dimension, and there is no sixth. The kill rule is hard: a composite score below four discards the idea automatically, and the kill screen has no try-again button.

One model scores; a second model checks it adversarially. Consensus is not guaranteed, and that is deliberate. Every assessment is stored with full provenance: model, prompt version, tokens in and out, and the raw output, with a per-row audit trigger recording who caused each change.

Local-first: your own Postgres, no telemetry, nothing leaves your infrastructure. A judgment you cannot reconstruct is a gut feeling with a better interface. Here every judgment can be reconstructed, down to the model and prompt that produced it.

SvelteKitPostgresDrizzleNode workerLISTEN/NOTIFY → SSE

Code available on request. Public release in preparation.

What is Idea Assessor?

Idea Assessor is a local-first tool that pressure-tests product ideas against a fixed YC-style rubric in exactly five rounds, with a hard kill rule and dual-model adversarial review, storing every assessment with full provenance.

Why only five rounds and a kill rule?

Because a constraint is what turns iteration into a decision. Five rounds are a budget only if there is no sixth; a kill rule disciplines only if it actually fires. Remove either and the tool becomes another notebook where judgments go to die.

Is the code available?

The tool is built and running locally. The code is available on request while the repository is being prepared for public release.